International Conference on Software Technology and Engineering (ICSTE 2012)
61 Unseen View Prediction Based on Single Image
Download citation file:
- Ris (Zotero)
- Reference Manager
Although great improvements have been made in face detection, it is still challenging to get reliable prediction of face views. The chief culprit is that the appearance of a face in a 2D image can change substantially as the viewing angles changes. In this Paper, a method for unseen view prediction based on single image is proposed. To realize the prediction of the unseen views, a 2D facial curve is utilized to describe for the 3D face model, and we analyze the projection relationship between the facial curve and the imaging plane, and got an effective method of transformation. The crucial techniques applied in the method were the foundation of analytic trigonometry and vector space model. The experimental results showed that with a facial curve single image could represent the face from a range of viewing angles(from -45 to 45 degree in clockwise direction), and the method was proved to be resultful with minor distortion.